Short video: TikTok, Reels and Shorts that stick
How TikTok, Instagram Reels and YouTube Shorts really decide what to show, what works for Duolingo and other real examples, and which tricks will cost you your account. With sources and honest evidence labels.
In short
TikTok, Instagram and YouTube all measure the same thing: does someone stop scrolling, how long do they watch and do they do something active (share, follow, search). Originality is now a hard gate: reused or template content isn't recommended or isn't paid out. Posting more often mainly increases your chance of an outlier, not your average.
- Intermediate
- 11 min
- free
- Fact-checked on 2 November 2026
- Completed
After this playbook you’ll know what the three big platforms really measure, what you can steal from accounts that do it well, and which tricks will cost you your account. Every number has a source and a label: ✅ primary (the platform or company itself) or ⚠ secondary/self-reported.
How it really works
Section titled “How it really works”What the platforms themselves say about their recommendations:
- TikTok ✅ (2020): looks at interactions (likes, shares, comments, follows), video info (captions, sounds, hashtags) and, with low weight, device and language. Watching a longer video all the way through “would receive greater weight”. Follower count and previous hits are not a direct factor.
- TikTok ✅: “unoriginal or reused material without anything new” doesn’t get into the For You feed.
- YouTube ✅: looks at whether people choose to watch, at average view duration and % viewed, then at likes and surveys.
- Instagram ✅ (2023): reels with a watermark, low resolution, no sound, with borders, mostly text, or that were already posted on Instagram are made less visible.
- Instagram ⚠ (Mosseri, Jan 2025, via trade press): three signals: watch time, likes per reach and sends per reach. Sends are “slightly more important” for reach among non-followers.
- YouTube ✅: “Disclosing AI content alone won’t limit a video’s audience or impact its eligibility to earn money.”
The common thread: (1) does someone stop scrolling, (2) how long do they watch, (3) do they do something active. And without originality you don’t get in.
Who does it perfectly
Section titled “Who does it perfectly”Duolingo (TikTok and others)
- What: mascot Duo behaves like a creator, not like an ad. Quick on trends, responds in other people’s comments. In February 2025 the “Dead Duo” storyline about the owl’s “death”.
- Result: according to its own shareholder letter, 1.7 billion organic impressions for the Dead Duo campaign (all channels combined), for which Duolingo “spent practically nothing” ✅ (SEC filing). Growth from 50,000 to over 16 million TikTok followers ⚠ (The Drum).
- Steal this: give your brand one recognisable character and a storyline that runs across several videos.
ZayaanFour (YouTube Shorts)
- What: relatable sketches in which the song lyric is the punchline, with audio from the library. That’s how he shows up on the sound pages.
- Result: more than 100 million views ⚠ (not verified ourselves).
- Steal this: let the audio do the work and pick a sound people are already searching for.
Khaby Lame (TikTok)
- What: wordless reactions to over-the-top “life hacks”: he simply does it the easy way, with a signature hand gesture.
- Result: in June 2022 the most followed TikTok account, around 142.8 million followers ⚠ (Euronews; current figure not checked).
- Steal this: one fixed, recognisable format that works without language.
Sabrina Ramonov (AI educator)
- What: recording one day a week: a YouTube tutorial, around 20 short videos and a newsletter. She uses her list of hooks as a lead magnet.
- Result: says she went from 0 to over 2.4 million followers in 24 months, solo and without ads ⚠ (self-reported; she also sells a repurposing tool).
- Steal this: batch your production on one fixed day instead of filming separately every day.
Do this
Section titled “Do this”1. Hook in second 0-1, three layers at once
Section titled “1. Hook in second 0-1, three layers at once”The first frame does three things: visual (movement or the end result), text (promise in max. 7 words) and voice (first sentence = the promise). YouTube looks at whether people choose to watch, TikTok weighs watching to the end heavily ✅.
- Write the last sentence (the payoff) first, then the hook that points to it.
- Start in the middle of the action. Cut all silence before the first word.
- Put the text hook in the safe zone, not at the bottom near the buttons.
Put the hook on screen, as on-screen text and in your first sentence at the same time. That way it also works with the sound off.
Free · runs entirely in your browser · nothing is stored or sent
2. Build on retention: beats and an open loop
Section titled “2. Build on retention: beats and an open loop”Average view duration and % viewed are official signals (YouTube ✅). Watch time and completion rate are in TikTok’s rewards formula ✅.
- Script in beats: hook → promise → 3 points → payoff. One beat = one shot.
- Put your strongest point last and announce it early (“the third is the only one that really works”).
- Read your retention graph: a dip is a spot where nothing happened. That’s where you cut next time.
3. Make every video searchable
Section titled “3. Make every video searchable”4. Make something people forward
Section titled “4. Make something people forward”5. Repurpose cleanly, and watch out for music on Shorts
Section titled “5. Repurpose cleanly, and watch out for music on Shorts”- Export the clean file from your editor. Never download from TikTok and upload to Reels: watermark = less visible ✅.
- Add text, audio and caption again per app.
6. Test with Trial Reels
Section titled “6. Test with Trial Reels”Instagram first shows a reel only to non-followers ✅ (since December 2024).
- Make 2-3 versions with one difference (only the hook, or only the audio).
- After 72 hours: compare views, skip rate and shares. Compare trial with trial, because trials often get less reach ⚠ (Mosseri).
- Share the winner with your followers.
Grey: possible, with risk
Section titled “Grey: possible, with risk”| Tactic | Risk | Safer |
|---|---|---|
| Reposting with light edits (crop, new music) | Instagram: already posted on IG = less visible ✅. TikTok: not in For You ✅. YouTube: “reused content” applies to the whole channel, even with permission ✅ | New hook and new recording. Other people’s work only via stitch/remix with your own commentary or a Collab post |
| Faceless AI channels at scale | YouTube “inauthentic content” (since July 2025): no money for content that is “mass-produced, generic, repetitive” or “heavily rel[ies] on generic templates” ✅ | Your own expertise and your own voice. Cloning your own voice doesn’t need to be labelled on YouTube ✅ |
| Comment bait (“type AI and I’ll send you the guide”) | TikTok: “Content that tries to manipulate people into giving gifts or inflating likes or follows isn’t eligible for the FYF” ✅. Simply asking is fine ✅ (YouTube) | Deliver value first, one CTA at the end, or a link in your bio |
| Mass posting via schedulers (3-10×/day) | TikTok prohibits “automation tools, scripts, or other tricks designed to bypass our systems” ✅, and the Creator Rewards formula doesn’t reward an “excessive amount of videos” ✅ | 3-5 unique posts per week via official schedulers |
Black: how it works and why you don’t do it
Section titled “Black: how it works and why you don’t do it”| Practice | What it is | Why it goes wrong |
|---|---|---|
| Bought views, followers, likes | Services deliver fake accounts or bot views | TikTok removes fake metrics and can restrict or ban you ✅. YouTube: up to channel removal, even if a hired agency did it ✅. Fake followers don’t watch, so your signals per reach drop |
| Engagement pods | Groups that like and comment on each other’s posts | Instagram prohibits “artificially collecting likes, followers, or shares” ✅. Your audience data gets polluted |
| Bots and view farms | Automated scrolling and watching | TikTok prohibits automation ✅. YouTube doesn’t count views from “automated click or scroll bots” for money ✅ |
| Sub4sub / follow-for-follow | Swap: I follow you, you follow me | YouTube explicitly prohibits it ✅ |
| Misleading deepfakes | Making a real person say something they didn’t say | TikTok prohibits it even with a label if it misleads on matters of public interest ✅ |
- “Post every day, or the algorithm punishes you.” Buffer: posting more raises the chance of an outlier, the median stays ~the same (489 → 459 views) ⚠.
- “Post at 7 pm on Tuesday.” Buffer (7.1 million posts) found the weekend slightly better, but says itself: “not all studies agree” ⚠. Your own Studio data wins.
- “Your follower count determines your reach.” TikTok says the opposite ✅.
- “The algorithm tests every video on 200-500 people first.” No official source found.
- “An AI label costs 50% reach.” No source. YouTube officially says it costs nothing ✅.
Measuring
Section titled “Measuring”Look at these five, not at views:
| What | Where | Tells you about |
|---|---|---|
| Swiping away in the first seconds | Instagram Skip Rate ⚠, YouTube “viewed vs swiped away” | your hook |
| Average watch time + retention graph | all three | where people drop off |
| Shares/sends per reach | Instagram, TikTok | reach among non-followers |
| Followers or leads per 1,000 views | all three | business value |
| Search terms | TikTok Creator Search Insights ✅, YouTube Analytics | lasting views |
Your first 30 days
Section titled “Your first 30 days”For a creator or freelancer in the AI niche:
- Day 1: pick one role as your target audience (freelancer, marketer, teacher) and put in your bio what they get from you.
- Day 2: note 20 search terms from Creator Search Insights and each platform’s autocomplete.
- Day 3: design one series format with a fixed opening and number (“AI tool of the week #1”). Each episode has its own topic.
- Week 1: batch 5 videos on one day: hook in three layers, one search term, subtitles.
- Week 1-4: post 3-5 times a week on all three, always the clean file.
- Every day: respond to serious comments in the first hour. Best question = next video.
- Week 2-3: test 2 hooks of the same video as a Trial Reel.
- Day 30: compare your 3 best and 3 worst videos on skip rate, watch time and shares per reach. Repeat what works, cut what sits below half your median.
Checklist
0/4
Your progress is only stored in this browser.
Test yourself
1What does Buffer's analysis of 11.4 million TikTok posts say about posting more often?
The gain is in the tail: more attempts, more chance of a hit. The median even went down slightly (489 to 459).
2You put a TikTok download with a watermark on Instagram Reels. What happens according to Instagram?
Instagram explicitly names watermarks as a reason to show reels less. Upload the clean file from your editor.
3Which faceless AI approach runs the biggest risk on YouTube?
Template-based mass production falls under 'inauthentic content' and isn't monetised. Cloning your own voice and AI subtitles are allowed.